Abstract: Based on a large depot project in Chongqing， the vibration and secondary noise characteristics and human comfort of the over-track buildings induced by metro operation in the depot are studied. The vibration characteristics of the depot are analyzed by field measurement. Combined with the numerical simulation method， the finite element model of the track-soil-depot-over-track buildings is established based on the metro-track coupling theory and the finite element theory. The vibration and secondary noise characteristics of the top buildings under train excitation are analyzed and evaluated. The combined annoyance model of vibration and secondary noise is constructed by combining psychology and fuzzy mathematics to analyze the human comfort， and the annoyance is used as the evaluation index. The results show that under the train operation， the vibration level of the throat area is the highest， and the vibration stability for different areas of the depot is as follows： the inner area > the throat area > the upper cover area. As the lateral propagation distance increases， the maximum Z vibration level of the platform decreases approximately linearly. The vibration peak of each building on the top is 12.5~20 Hz， and the main frequency band of secondary noise is 63~80 Hz. The vibration and secondary noise of each building attenuate with the increase of floors， but the vibration and secondary noise of residential building are amplified after 12 floors. The vibration of each building on the upper cover do not exceed the standard， but the secondary noise exceeds the standard in the commercial building， and the maximum exceeding value is 4.9 dB（A）. The annoyance results obtained by the joint annoyance model are in good agreement with the standard evaluation results， but the annoyance can refine the influence of vibration and secondary noise on human comfort. The model calculation shows that although the first floor of the residential building meets the standard limit， the joint annoyance is above 0.6 at night， and the evaluation using the joint annoyance rate model is more demanding.